“Artificial intelligence companies have run out of data for training their models and have “exhausted” the sum of human knowledge. The ‘only way’ to counter the lack of source material for training new models was to move to synthetic data created by AI.” (Source: Guardian via News Items.)
So it turns out that taking all of the music in the world and using it to create software that can “create” great “new” music for people is expensive and hard to do. So much so that AI companies mostly don’t want to do it. It’s also maddeningly tricky and time-consuming to license music, especially publishers. This applies to other mediums, such as print, video, etc., but music is excruciatingly slow.
So, if you are an AI company, your dream is to be able to do what you do without any of the headaches, costs, and time constraints imposed by those pesky creators with their “shortsighted” petitions and lawsuits.
This is why the holy grail for AI founders and engineers is to create their own data (based on work taken from other people, naturally) and never need to train on anyone else's work again. This is what they refer to as “synthetic data.”
The Synthetic Data Dream: Mimicking Artists Without Their Input
“In essence, synthetic data is any data produced algorithmically rather than captured from real-world events.” Dr Tamay Aykut - CEO Sureel AI
Today, and for many years now, AI companies are heaving enormous sums of venture capitalist's money into models that can generate new music indistinguishable in its sound, feel, and composition from our current and past human-created music. Their goal is clear: to create songs that sound like our favorite artists without using a single note of their original music. This approach relies on the aforementioned synthetic data. If the goal is reached, it could render original artist contributions, licenses, and royalties completely unnecessary. Today, this dream, while tantalizingly close and within sight, is still extremely challenging. However, with the current flood of investment and the brilliant minds set to achieve this mission, it will likely be a success before we know it.
The Race Against Time: Why Proper Licensing is Critical Now
The urgency for artists and rightsholders to secure proper licensing agreements has never been greater. As AI technology advances along its seeming exponential curve, the window for establishing fair compensation and attribution is rapidly closing. Artists risk losing control over their musical legacy and future earnings without immediate action, and the means of licensing are critical to creators owning a dignified piece of the Attribution share future.
What Does an AI Want?
“Show me the incentive, and I'll show you the outcome" - Charlie Munger
The technology underpinning AI represents an unprecedented opportunity for humanity and music creation. Still, it also poses a significant risk to artists' livelihoods. While some IP owners have voluntarily participated, albeit with short-sighted and potentially catastrophic consequences to their businesses, others remain cautious about AI's role in the fast-emerging race to get ahead of their competitors as they should. Worst case scenario: IP rightsholders receive money from AI companies in exchange for their files and promises that whilst said firms have not done the right thing up until now, they intend to do so once sanctioned and therefore blessed. In other words, trust us, we’ll account and attribute for you. You can trust us, we only needed to take your music that one time; we are the vitamin, not the virus.
But as Musk, quoted above, along with others building these models are now implying: the world's knowledge (unlicensed and uncompensated) has already been “exhausted” to get this far. In that case, synthetic data (created from the first theft) will be needed to save the day. Therefore any license that does not take into account this reality will be a nail in the coffin for one of the most significant potential revenue streams most creators will ever see. A catastrophic future for the music industry is not inevitable, but if our MP3/Napster past is anything to go by, it becomes more likely by the day. Any promise of self-created “guard rails” or self-calculated attribution to rightsholders they’ve already taken from is, to say the least, short-sighted. This technological problem requires a technological solution, backed up by a legal standard that starts with the creator first. We must continue to respect our artists and human created music through a music licensing model that is monitored by neutral third-party attribution. This is the only way to ensure that creators can grow with these multi-billion dollar AI companies. As opposed to what we are doing now, which is simply watching them grow as we wring our hands, sign petitions, or wait it out.
As a simple guide, we have published the following Music Rights Holders Checklist for Licensing AI Companies
